Industrial vision-based myopia-prevention paper quality detection system

Abstract

The present application relates to the technical field of industrial vision detection, in particular to a myopia-preventing paper quality detection system based on industrial vision, comprising: a multi-angle image acquisition module for acquiring a paper surface gray image under multi-angle illumination environment; a frequency domain analysis module for analyzing frequency distribution characteristics and calculating local spectrum entropy to quantify diffuse reflection randomness; a texture analysis module for constructing a gray level co-occurrence matrix to extract texture energy difference degree and represent light reflection isotropic characteristics; a glossiness analysis module for calculating glossiness consistency index to evaluate light softness degree; and a defect identification module for fusing local spectrum entropy, texture energy difference degree and glossiness consistency index to construct a multi-dimensional texture feature set, identifying functional diffuse reflection abnormal areas and generating a detection result; the present application quantifies visual comfort from a microscopic optical perspective, ensuring functional quality of light softness of myopia-preventing paper.

Description

Technical Field

[0001] This invention relates to the field of industrial vision inspection technology, specifically to an anti-myopia paper quality inspection system based on industrial vision. Background Technology

[0002] In the field of anti-myopia paper production and manufacturing, the core quality attributes of paper lie in its surface microstructure's ability to diffusely reflect light and its visual comfort. Existing industrial vision inspection solutions generally adopt physical defect detection modes based on geometric morphology analysis, mainly focusing on obvious physical damage such as holes, dirt spots, or edge damage. Although this solution can identify macroscopic structural anomalies, it is difficult to distinguish between the inherent high-frequency random texture of pulp fibers and functional defects caused by uneven coating processes due to the lack of quantitative means for microscopic optical properties. Existing technologies cannot effectively evaluate key indicators such as local spectral entropy, texture energy anisotropy, and gloss consistency, resulting in the inability to identify specular reflection and directional glare areas caused by excessive calendering or coating agglomeration. In addition, traditional detection methods mainly rely on fixed thresholds or manual qualitative judgment, which is difficult to adapt to the dynamic drift of papermaking processes and is prone to misjudgment due to background noise interference, failing to guarantee the functional quality of anti-myopia paper with soft light. Therefore, how to establish a multi-dimensional feature system for quantifying visual comfort from a microscopic optics perspective, and how to solve the problem of accurate positioning and adaptive detection of functional diffuse reflection abnormal areas in complex texture backgrounds, have become urgent technical problems to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an anti-myopia paper quality inspection system based on industrial vision. Specifically, the technical solution of this invention includes: The multi-angle image acquisition module is used to perform grayscale imaging on the surface of anti-myopia paper under multi-angle lighting conditions to obtain grayscale images of the paper surface. The frequency domain analysis module is used to perform local frequency domain transformation on the grayscale image of the paper surface, analyze the frequency distribution characteristics, and calculate the local spectral entropy based on the frequency distribution characteristics to quantify the diffuse reflectance randomness of the microstructure of the paper surface. The texture analysis module is used to perform spatial domain texture statistics on grayscale images of paper surfaces, construct a gray-level co-occurrence matrix, and extract texture energy difference based on the gray-level co-occurrence matrix to characterize the isotropic characteristics of light reflection on the paper surface. The gloss analysis module is used to perform statistical analysis on the pixel grayscale distribution of grayscale images of paper surfaces and calculate the gloss consistency index to evaluate the softness of light on the paper surface under simulated human eye perception. The defect identification module is used to fuse local spectral entropy, texture energy difference, and gloss consistency index to construct a multi-dimensional texture feature set, and to identify functional diffuse reflection abnormal areas based on the multi-dimensional texture feature set to generate quality inspection results.

[0004] Preferably, the method for obtaining a grayscale image of the paper surface includes: The lighting unit is controlled to illuminate the paper surface sequentially at multiple preset incident angles; At each incident angle, the imaging unit is triggered to expose, and multiple frames of raw illumination images are acquired. Weighted fusion processing is performed on multiple frames of original illumination images to suppress high-frequency random noise and generate a grayscale image of the paper surface.

[0005] Preferably, the method for calculating local spectral entropy includes: The grayscale image of the paper surface is divided into multiple overlapping local analysis windows; Perform a two-dimensional discrete Fourier transform on the image data within each local analysis window to obtain the local spectral amplitude matrix; The local spectral amplitude matrix is ​​normalized to obtain the normalized spectral probability distribution; Based on the normalized spectral probability distribution, the information entropy value is calculated, and the information entropy value is used as the local spectral entropy corresponding to the local analysis window.

[0006] Preferred methods for extracting texture energy differences include: Based on the grayscale image of the paper surface, multiple pre-selected directional offsets are set; For each directional offset, a corresponding gray-level co-occurrence matrix is ​​constructed. For each gray-level co-occurrence matrix, calculate the quadratic angular moment eigenvalues ​​as the texture energy values ​​for the corresponding directions; Calculate the standard deviation of the texture energy values ​​in all directions, and use the standard deviation as the texture energy difference.

[0007] Preferably, the method for calculating the gloss consistency index includes: Extract the pixel grayscale histogram of the grayscale image of the paper surface; Calculate the mean, standard deviation, and skewness coefficient of the pixel grayscale histogram; Based on preset weighting coefficients, the standard deviation and skewness coefficient are weighted and summed to obtain the gloss consistency index, where the skewness coefficient is used to characterize the distribution tendency of local highlight areas.

[0008] Preferably, methods for identifying regions of functional diffuse aberration include: Generate an entropy mapping map and an energy difference mapping map with the same size as the grayscale image of the paper surface; Traverse the entropy value mapping map and filter out areas where the local spectral entropy is lower than the preset diffuse reflection threshold, marking them as suspected specular reflection areas; Traverse the energy difference map and filter out areas where the texture energy difference is higher than the preset isotropic threshold, marking them as suspected coating unevenness areas; The intersection of the suspected specular reflection area and the suspected coating unevenness area is taken, and the intersection area is regarded as the functional diffuse reflection anomaly area.

[0009] Preferably, the method for generating quality inspection results includes: The percentage area of ​​functional diffuse reflectance anomalous regions in grayscale images of paper surfaces; Obtain the average gloss consistency index within the functional diffuse reflection anomaly area; A preset quality weight set is provided, which includes area weight and gloss weight. Based on the quality weight set, the visual comfort score is calculated by weighted summation of the area ratio and the average value. The visual comfort score is compared with a preset pass / fail benchmark score: if the visual comfort score is lower than the pass / fail benchmark score, a failure instruction is generated as the quality inspection result; if the visual comfort score is not lower than the pass / fail benchmark score, a pass instruction is generated as the quality inspection result.

[0010] Preferably, it also includes an adaptive feedback adjustment module, used for: Obtain feature data of functional diffuse reflection anomaly regions corresponding to all non-compliant instructions generated within a historical time period; Cluster analysis was performed on the feature data to identify the cluster centers of the main defect types; Based on the data distribution characteristics of the cluster centers, the preset diffuse reflection threshold and preset isotropic threshold are updated to correct the system's sensitivity to specific process fluctuations.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention analyzes the local spectral entropy by frequency domain transformation and extracts the texture energy difference by constructing a gray-level co-occurrence matrix. It quantifies the randomness of diffuse reflection and the isotropicity of light reflection on the paper surface at the microscopic level. This breaks through the limitation of traditional machine vision, which can only detect geometric defects such as holes. It effectively identifies functional abnormalities such as glare or directional reflection caused by uneven coating process, thereby ensuring the soft visual comfort of anti-myopia paper. 2. This invention employs a multi-frame image weighted fusion processing technique under multi-angle lighting conditions, which effectively suppresses high-frequency random noise generated by the pulp fibers themselves, while enhancing the low-frequency characteristics of uneven coating areas. This method significantly improves the image signal-to-noise ratio, avoids interference from the inherent texture background of the paper on the detection algorithm, and ensures that the system can accurately capture weak diffuse reflection coating defects in complex industrial environments. 3. This invention calculates the gloss consistency index by statistically analyzing the skewness coefficient of pixel grayscale distribution, and calculates the visual comfort score by combining area and gloss weight, simulating the human eye's perception characteristics of light intensity distribution and local highlights; this not only solves the problem of the difficulty in unifying subjective quality inspection standards, but also can keenly eliminate defective products that are qualified in overall brightness but have local glaring bright spots, thus realizing an objective quantitative evaluation of the anti-myopia function indicators. 4. The present invention sets up an adaptive feedback adjustment module, which identifies the main defect patterns by clustering analysis of historical non-conforming feature data, and dynamically updates the diffuse reflection threshold and isotropic threshold accordingly. This mechanism corrects the system's sensitivity to specific process fluctuations, enabling it to automatically adapt to changes in raw material batches or production environment, effectively reducing the false alarm rate while maintaining strict quality bottom lines. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 The anti-myopia paper quality inspection system based on industrial vision includes: a multi-angle image acquisition module, which is used to perform grayscale imaging on the surface of the anti-myopia paper under multi-angle lighting conditions to obtain grayscale images of the paper surface; The frequency domain analysis module is used to perform local frequency domain transformation on the grayscale image of the paper surface, analyze the frequency distribution characteristics, and calculate the local spectral entropy based on the frequency distribution characteristics to quantify the diffuse reflectance randomness of the microstructure of the paper surface. The texture analysis module is used to perform spatial domain texture statistics on grayscale images of paper surfaces, construct a gray-level co-occurrence matrix, and extract texture energy difference based on the gray-level co-occurrence matrix to characterize the isotropic characteristics of light reflection on the paper surface. The gloss analysis module is used to perform statistical analysis on the pixel grayscale distribution of grayscale images of paper surfaces and calculate the gloss consistency index to evaluate the softness of light on the paper surface under simulated human eye perception. The defect identification module is used to fuse local spectral entropy, texture energy difference, and gloss consistency index to construct a multi-dimensional texture feature set, and to identify functional diffuse reflection abnormal areas based on the multi-dimensional texture feature set to generate quality inspection results.

[0015] This embodiment details the core architecture of the system, aiming to solve the technical problem that existing technologies only focus on physical defects while neglecting the core functional attributes of anti-myopia paper; the system is equipped with a multi-angle image acquisition module, which uses a ring multi-angle light source array to acquire raw data that can reflect the optical properties of the paper surface micromorphology; a frequency domain analysis module is involved to quantify the micro-disorder of the paper surface from the perspective of frequency distribution; The local spectral entropy calculated by this module refers to the degree of disorder in the energy distribution of a local region of the image in the frequency domain. The higher the entropy value, the more disordered the microstructure and the better the diffuse reflection effect. The texture analysis module evaluates the directional characteristics of the paper surface to reflect light. It extracts the texture energy difference by constructing a gray-level co-occurrence matrix. This difference represents the standard deviation of the texture energy characteristics in different directions. The lower the difference, the more consistent the reflective properties of the paper in all directions. Based on this, the gloss analysis module simulates the human eye's perception of light intensity distribution and calculates the gloss consistency index; the defect identification module executes feature fusion logic, specifically by merging the spatially aligned local spectral entropy. Texture energy difference and gloss consistency index Perform channel cascading, constructing a dimension of The multidimensional texture feature tensor; in, and These represent the height and width pixel dimensions of the grayscale image on the paper surface, respectively. Based on this feature tensor, the system accurately locates the regions affecting visual function and generates quality detection results. This embodiment, by introducing frequency domain entropy and texture isotropic analysis, breaks through the limitation of traditional machine vision that can only detect geometric defects, and quantifies the visual comfort of anti-myopia paper from the perspective of microscopic optics. The system effectively eliminates those unqualified products that, although they have no physical damage, cause local glare due to uneven coating process, thus ensuring the functional quality of anti-myopia paper with soft light at the source of production.

[0016] Example 2: The method for acquiring a grayscale image of a paper surface includes: controlling the illumination unit to sequentially illuminate the paper surface at multiple preset incident angles; triggering the imaging unit to expose at each incident angle to acquire multiple frames of original illumination images; performing weighted fusion processing on the multiple frames of original illumination images to suppress high-frequency random noise and generate a grayscale image of the paper surface.

[0017] This embodiment further defines the execution logic of the multi-angle image acquisition module, aiming to overcome the interference of high-frequency random noise of the pulp fiber itself on coating defect detection; the system controls the illumination unit to illuminate the paper surface sequentially at multiple preset incident angles, such as 30 degrees, 45 degrees and 60 degrees; at each incident angle, the imaging unit is triggered to expose and acquire multiple frames of original illumination images. In this step, considering that the continuous movement of paper on the industrial production line may cause pixel displacement between images at different angles, the system performs image registration before fusion: the FFT phase correlation algorithm is used to calculate the translation vector of each frame image relative to the first frame image. And apply affine transformation to align all images to the same pixel coordinate system; To address the technical issue of edge artifacts caused by zero-fill regions introduced at the edges of images after affine transformation during weighted fusion, the system adds an effective field-of-view intersection step before executing the fusion formula: for each frame of aligned image... Generate a binary validity mask of the same size. The original pixel region is marked as 1, and the filled region is marked as 0; the intersection of all angle masks is calculated using the following formula: ,based on Extract the boundary coordinates of the largest inscribed rectangle and all Clip to the unified valid coordinate range; The system introduces a multi-angle fusion algorithm to perform weighted fusion processing on the cropped multiple frames of the original illumination images. The calculation formula is as follows: , in, The pixel grayscale values ​​of the generated grayscale image of the paper surface are derived from calculations and have the physical meaning of being the fused low-noise image signal. This represents the preset total number of incident angles, sourced from system settings, and physically signifies the coverage dimension of the sampling angles. Subscripts are used here to distinguish them and avoid confusion with subsequent image size symbols. The original illumination image at the k-th incident angle contains registered and cropped pixel grayscale values, sourced from the imaging unit, and physically representing the optical response at a single angle. The fusion weight coefficient corresponds to the k-th angle. It is a preset value, and its physical meaning is the contribution of illumination from different angles to the final image. To ensure that the dynamic range of the fused image's brightness does not overflow, the preset weight coefficient must satisfy a normalization condition, i.e. For example, the values ​​can be 0.3, 0.4, and 0.3 at the three angles respectively; This embodiment utilizes the relative stability of the diffuse reflection coating response at different angles and the high-frequency fluctuation of the pulp fiber noise response at different angles to achieve physical-level low-pass filtering through multi-angle weighted fusion. This method effectively suppresses background interference from paper fiber texture while significantly enhancing the low-frequency characteristics of the uneven coating region, thereby improving the signal-to-noise ratio of subsequent algorithms for detecting diffuse reflection anomalies.

[0018] Example 3: The method for calculating local spectral entropy includes: dividing the grayscale image of the paper surface into multiple overlapping local analysis windows; performing a two-dimensional discrete Fourier transform on the image data within each local analysis window to obtain the local spectral amplitude matrix; normalizing the local spectral amplitude matrix to obtain the normalized spectral probability distribution; calculating the information entropy value based on the normalized spectral probability distribution, and using the information entropy value as the local spectral entropy corresponding to the local analysis window.

[0019] This embodiment details the specific steps of the frequency domain analysis module to quantify the diffuse reflectance capability of microstructures; in order to ensure that the size of the subsequently generated feature map is strictly consistent with the original image, that is, to satisfy the pixel-level correspondence, the system performs edge filling on the grayscale image of the paper surface; A reflection-fill strategy is used, with the fill width being half the side length of the local analysis window. To avoid boundary frequency artifacts caused by zero padding, the padded image is divided into multiple overlapping local analysis windows, such as 32 by 32 pixels, with a sliding step size of 1 pixel. A two-dimensional discrete Fourier transform is performed on the image data within each local analysis window to obtain the local spectral amplitude matrix. In order to eliminate the interference of the background average brightness on the texture analysis, the DC component in the spectral amplitude matrix, that is, the component with a frequency of 0, is set to zero at this time. The spectral amplitude is converted into a probability distribution form to obtain the normalized spectral probability distribution. In this step, the system calculates the sum of the spectral amplitudes. ,like Smaller than a tiny amount ,like If the window entropy is zero, set it to 0 to prevent division by zero errors; otherwise, perform normalization calculations as follows: , Calculating local spectral entropy based on information theory principles The entropy value is then assigned to the center pixel of the window to form a local spectral entropy map, calculated using the following formula: , When implementing this formula in the code, the system introduced numerical stability protection logic: for any frequency component, if Then directly the item The value is determined to be 0, following the rules. The mathematical limit is used to prevent computer programs from throwing non-numeric values, i.e., non-numeric exceptions; in, The local spectral entropy corresponds to the local analysis window. It is calculated and its physical meaning is the disorder of image texture in the frequency domain, with the unit being bits. The pixel size of the local analysis window in the horizontal and vertical directions is derived from system settings and physically represents the sampling space range of the frequency domain transform. These are the horizontal and vertical coordinate indices in the frequency domain, derived from the DFT transform results, and physically representing the positions of the frequency components. The normalized spectral probability distribution is derived from the normalization of the local spectral amplitude matrix, and its physical meaning is the energy proportion of a specific frequency component. This embodiment uses local spectral entropy as an indicator to measure the disorder of image texture, accurately distinguishing between diffuse reflection areas and specular reflection defect areas. For high-quality anti-myopia paper, the high randomness of its surface microstructure is manifested in the frequency domain as a uniform energy distribution, i.e., a high entropy value, while specular reflection areas caused by excessive calendering or coating agglomeration are manifested as low entropy values. Thus, from the essence of physical optics, a quantitative assessment of the quality of the paper surface microstructure is achieved.

[0020] Example 4: The method for extracting texture energy difference includes: setting multiple pre-selected directional offsets based on the grayscale image of the paper surface; constructing a corresponding grayscale co-occurrence matrix for each directional offset; calculating the quadratic angular moment eigenvalue for each grayscale co-occurrence matrix as the texture energy value for the corresponding direction; calculating the standard deviation of the texture energy values ​​for all directions, and using the standard deviation as the texture energy difference.

[0021] This embodiment details the specific steps of the texture analysis module in detecting the isotropic light reflection of the paper surface; a sliding window, such as 32x32 pixels with a step size of 1 pixel, is used to traverse the grayscale image of the paper surface and set multiple pre-selected directional offsets, such as 0 degrees, 45 degrees, 90 degrees and 135 degrees, which are aligned with the frequency domain analysis module. For each directional offset, a gray-level co-occurrence matrix corresponding to the image data within the current window is constructed. During the construction process, to balance computational efficiency and texture capture capability, a distance step size for pixel pairs is set. The image is compressed to 16 gray levels, with each gray-level co-occurrence matrix represented by 1 pixel. The second moment is calculated for each gray-level co-occurrence matrix using the following formula: , in, These are the row and column indices of the gray-level co-occurrence matrix, derived from the matrix dimension definition. Physically, they represent the gray-level values ​​of the reference pixel and neighboring pixels, respectively, and their values ​​range from [value range missing]. In this embodiment ; For the first The elements of the normalized gray-level co-occurrence matrix in each direction represent the probability of a specific gray-level combination occurring. The quadratic moment value is used as the texture energy value in that direction; the standard deviation of the energy values ​​in all directions is calculated to obtain the texture energy difference D of the center pixel of the window. After traversal, a texture energy difference map is generated. The calculation formula is as follows: , in, The texture energy difference is derived from calculation and its physical meaning is the directional sensitivity of the texture features on the paper surface; it is dimensionless. This represents the total number of pre-selected directional offsets, which are derived from system settings and physically represent the angular resolution of texture analysis. For the first The texture energy value in each direction is derived from the second-order moment calculation of the gray-level co-occurrence matrix, and its physical meaning is the uniformity of image gray level in that direction; It is the arithmetic mean of texture energy values ​​in all directions, derived from statistical calculations, and its physical meaning is the baseline level of overall texture energy; This embodiment effectively identifies anisotropic reflective defects that cause visual fatigue by calculating the texture energy difference. Ideal anti-myopia paper should have isotropic diffuse reflection characteristics, that is, the difference should be close to zero. However, coating scratches or directional calendering defects can cause abrupt changes in texture energy in a specific direction, which significantly increases the difference. This indicator ensures the consistency of visual experience of the paper under various viewing angles.

[0022] Example 5: The method for calculating the gloss consistency index includes: extracting the pixel gray-level histogram of the gray-level image of the paper surface; calculating the mean, standard deviation, and skewness coefficient of the pixel gray-level histogram; and calculating the gloss consistency index by weighted summation of the standard deviation and skewness coefficient based on preset weighting coefficients, wherein the skewness coefficient is used to characterize the distribution tendency of local highlight areas.

[0023] This embodiment details the specific steps of the gloss analysis module in evaluating the softness of light. To enable subsequent refined quantification of abnormal areas, i.e., to support the calculation of the average value of the region in Embodiment 7, this embodiment uses a local sliding window mechanism to generate a gloss feature map, rather than calculating only a single index for the entire image. The system defines the size as... ,in, The number of pixels representing the side length of the sliding window, for example A sliding window of pixels is used to traverse the grayscale image on the paper surface with a step size of 1; for each window position, the grayscale histogram of the pixels within the window is extracted; the mean, standard deviation and skewness coefficient of the local histogram are calculated. In this step, to pass code-level robustness testing, the system introduces numerical protection logic for solid color backgrounds or images with extremely low contrast: the standard deviation is determined before calculating the skewness coefficient. Is it less than a preset machine accuracy threshold, for example? If the value is less than this threshold, it indicates that the image lacks grayscale variation, and the skewness coefficient is directly adjusted. Assign a value of 0 to avoid division by zero anomalies; if the standard deviation is within the normal valid range, then apply the skewness coefficient. The calculation formula is: , in, The total number of pixels within the window. For the first grayscale value of each pixel. The mean, The standard deviation is used to eliminate the difference in magnitude between the standard deviation and the skewness coefficient (i.e., the dimensionless statistical moment), which depends on the gray level range (e.g., 0-255). To ensure the effectiveness of the weighted calculation, a maximum gray level constant is introduced to normalize the standard deviation. Based on preset weighting coefficients, the normalized standard deviation and the absolute value of the skewness coefficient are weighted and summed to obtain the gloss consistency index corresponding to the center pixel of the window. The calculation formula is as follows: , in, coordinates The gloss consistency index at the location is used to ultimately construct a gloss index mapping map with the same size as the original image; the smaller the value, the softer the image. The standard deviation of the local pixel grayscale histogram; This is the maximum gray level value of the image; for example, 255 is used for an 8-bit depth image. The skewness coefficient of the local pixel grayscale histogram; For example, take the preset weighting coefficients. ; The weighting coefficients were obtained as follows: a set of anti-myopia paper samples with different gloss characteristics were collected in advance, and multiple ophthalmologists were invited to score visual comfort from 0 to 10 points to establish a subjective score dataset of the samples; the correlation between the normalized standard deviation, skewness coefficient and subjective scores of the samples was analyzed using a multiple linear regression algorithm, and the coefficients of the regression equation were used as the weighting coefficients. and The optimal solution is obtained to ensure that the calculated index can objectively reflect the subjective perception of the human eye. Regarding the physical meaning of the weighted summation in the formula, this embodiment further explains: Although the gray value of an image pixel is essentially a dimensionless relative light intensity value, directly using the unnormalized standard deviation will cause its value to be much larger than the skewness coefficient. Divide by the maximum gray level , making Become the skewness coefficient Normalized indices within similar numerical ranges ensure the mathematical validity of linear superposition; therefore, the gloss consistency index... Defined as a dimensionless scalar reflecting relative intensity, the coefficients are obtained through regression. and It is also set as a dimensionless weighting factor; Specifically, the formula uses the absolute value of the skewness coefficient. This is because both positive bias (local highlight concentration) and negative bias (local shadow) represent uneven gloss distribution, deviating from the ideal soft diffuse reflection state, i.e., normal distribution. This treatment ensures that the index has consistent sensitivity to various types of non-uniform gloss. This embodiment combines standard deviation and skewness coefficient to comprehensively evaluate whether the paper surface meets the optical standards for myopia prevention, which are characterized by soft light and no glaring highlights. In particular, the introduction of skewness coefficient can accurately capture local highlight areas, i.e., specular reflection points, that cause the histogram to show a long tail distribution. This simulates human eye perception and accurately eliminates defective products that, although the overall brightness is qualified, have local glaring highlights.

[0024] Example 6: The method for identifying functional diffuse reflection anomalous regions includes: generating an entropy map and an energy difference map with the same size as the grayscale image of the paper surface; traversing the entropy map to filter out regions with local spectral entropy lower than a preset diffuse reflection threshold and marking them as suspected specular reflection regions; traversing the energy difference map to filter out regions with texture energy difference higher than a preset isotropic threshold and marking them as suspected coating unevenness regions; taking the intersection of the suspected specular reflection regions and the suspected coating unevenness regions, and using the intersection region as the functional diffuse reflection anomalous region.

[0025] This embodiment details the specific steps of the defect identification module to accurately locate defects, and adopts an intersection verification strategy; the system generates an entropy value mapping map and an energy difference mapping map with the same size as the grayscale image of the paper surface, where each pixel corresponds to the entropy value and energy difference degree of its window; The entropy map is iterated, and regions with local spectral entropy below a preset diffuse reflection threshold are marked as suspected specular reflection areas. Simultaneously, the energy difference map is iterated, and regions with texture energy difference above a preset isotropic threshold are marked as suspected coating unevenness areas. The intersection of the suspected specular reflection areas and the suspected coating unevenness areas is used to define the functional diffuse reflection anomaly region. The logic is as follows: , in, coordinates The local spectral entropy value at a point is derived from an entropy mapping map, and its physical meaning is the intensity of diffuse reflection randomness of the microstructure in the neighborhood of that point. coordinates The texture energy difference value at a point is derived from the energy difference map, and its physical meaning is the degree of directional anisotropy of the texture in the neighborhood of that point. The preset diffuse reflection threshold is obtained by pre-collecting data. A sample of qualified paper was analyzed, and the mean value of its local spectral entropy was calculated. with standard deviation The calculation formula is as follows: , The physical meaning is the minimum tolerance limit for the randomness of diffuse reflection; To preset the isotropic threshold, the mean value of the texture energy difference of qualified samples is similarly calculated. with standard deviation ,set up The physical meaning is the maximum tolerance limit for differences in texture direction; These are pixel coordinates, originating from the image spatial domain, and physically representing defect location points. This embodiment greatly reduces the false detection rate caused by background noise through intersection judgment logic. Since functional defects such as glare caused by uneven coating must have both low diffuse reflection randomness and high directional sensitivity, this method avoids misjudgment caused by single features due to changes in paper fiber density or tiny wrinkles, ensuring that the identified area is indeed an optical defect that affects the visual experience.

[0026] Example 7: The method for generating quality inspection results includes: statistically analyzing the proportion of functional diffuse reflectance aberration regions in the grayscale image of the paper surface; obtaining the average gloss consistency index of functional diffuse reflectance aberration regions; pre-setting a quality weight set, which includes area weight and gloss weight; and calculating the visual comfort score by weighted summation of the proportion of area and average value based on the quality weight set. The visual comfort score is compared with a preset pass / fail benchmark score: if the visual comfort score is lower than the pass / fail benchmark score, a failure instruction is generated as the quality inspection result; if the visual comfort score is not lower than the pass / fail benchmark score, a pass instruction is generated as the quality inspection result.

[0027] This embodiment details the computational logic for converting multidimensional physical features into quantitative quality indicators, aiming to solve the technical problem of the difficulty in standardizing qualitative descriptions in traditional quality inspection; the system traverses the binarized functional diffuse reflection anomaly region mask and counts the total number of pixels. and divide by the total number of pixels in the image. Multiply by the constant 100 to get the area percentage, expressed as a percentage (%), with a value ranging from 0 to 100. The formula is as follows: , Based on the gloss consistency index mapping generated in Example 5 Obtain the exponent values ​​corresponding to all pixels within the mask of the functional diffuse reflection anomaly region, and calculate their arithmetic mean. If the area is Then directly order To avoid calculation errors; The system calls the preset set of quality weights. ,in, This is the area penalty weight, for example, a value of 2.0, which physically means that 2 points are deducted for every 1% of the defect area. This is the gloss penalty weight, for example, a value of 15.0, which physically means the deduction range corresponding to a unit gloss index; Weighting coefficient and The specific acquisition method is as follows: Select M groups, for example, M=200, of representative defect samples. A group of senior quality inspection experts will score the visual comfort of each group of samples, with a maximum score of 100, to obtain the subjective score set. Simultaneously calculate the defect area ratio for each sample group. and average gloss index Construct a linear regression equation, the calculation formula of which is: , Solving the equation using the least squares method yields regression coefficients that represent the optimal weighting coefficients. This process ensures consistency between machine scoring standards and human expert evaluation standards; a relatively large weighting coefficient is set here. In order to and Dimensional normalization was performed to balance the order-of-magnitude difference between the gloss index and the area ratio; visual comfort scores were calculated. The scoring model uses a base score deduction method, and the calculation formula is as follows: , The part in parentheses In this example, the weighted sum of the area ratio and the average value is used to calculate the total visual discomfort score. Subtracting this from the maximum score of 100 yields the visual comfort score. This formula introduces... The function acts as a boundary protection to prevent negative scores due to severe defects; the calculated... Compare the score with a preset passing score, such as 85 points: If If the visual comfort level is deemed insufficient, an unqualified instruction is generated; otherwise, a qualified instruction is generated. This embodiment achieves a refined grading of the quality of anti-myopia paper by constructing a weighted scoring model that includes a dimensional balance factor. This method not only considers the geometric scale of the defect, i.e., the area, but also amplifies the penalty for high reflectivity by using gloss weight, ensuring that even laser-like defects with small defect areas but extremely glaring reflectivity can be effectively intercepted.

[0028] Example 8: The adaptive feedback adjustment module is used to: acquire feature data of functional diffuse reflection anomaly regions corresponding to all non-compliant instructions generated within a historical time period; perform cluster analysis on the feature data to identify the cluster centers of the main defect types; Based on the data distribution characteristics of the cluster centers, the preset diffuse reflection threshold and preset isotropic threshold are updated to correct the system's sensitivity to specific process fluctuations.

[0029] This embodiment details the closed-loop control algorithm of the adaptive feedback adjustment module, aiming to solve the problem that a fixed threshold cannot adapt to the dynamic drift of the papermaking process. The system extracts feature data of all abnormal areas judged as unqualified within a historical time period, such as the most recent 24 hours, from the database, and constructs a feature vector set, the calculation formula of which is: , To address the feature space distortion caused by the inconsistency in physical dimensions, specifically the local spectral entropy (in bits) and texture energy difference (dimensionless), the system performs feature dimension normalization: it calls pre-stored statistical parameters from the golden template, namely the mean. with standard deviation , the original feature data Transformed into a dimensionless standard feature vector, its dimension is 2, corresponding to and The normalized value is calculated using the following formula: , This step eliminates the unexpected interference of differences in the magnitude of different features on the weights of subsequent algorithms, ensuring the consistency of physical meaning; cluster analysis is performed on the normalized feature data, specifically using the K-Means algorithm or the DBSCAN density clustering algorithm, to identify the cluster centers of the main defect types; For isotropic threshold and diffuse reflection threshold The system calculates the target update value. The calculation formula is as follows: , , in, and These represent the mean and standard deviation of the identified defect cluster centers along the local spectral entropy dimension, respectively. and These are the mean and standard deviation in the dimension of texture energy difference, respectively; the calculation logic aims to move the threshold boundary to a statistical lower or upper limit that covers the current main process fluctuation range; To prevent the system from misclassifying real physical defects as normal process fluctuations due to blindly following substandard data (i.e., to avoid threshold drift risk), the system introduces a property discrimination logic based on Mahalanobis distance before updates: The normalized feature distribution parameters of the gold sample, stored in read-only memory, are pre-loaded and denoted as the mean vector. The theoretical value is the zero vector, and the covariance matrix. That is, the correlation coefficient matrix of features; calculate the currently identified defect cluster centers. Mahalanobis distance from the gold standard distribution The calculation formula is as follows: , in, The unit is dimensionless and is used for determining statistical significance. Since the input vector has been dimensionless normalized, the result calculated here is... Essentially the square of the Mahalanobis distance, it possesses clear statistical and physical significance, accurately characterizing the statistical significance of the current defect mode deviating from the normal process baseline; a drift judgment threshold is set. For example, the value is 16.0; threshold The basis for obtaining the threshold is statistical boundary theory: since the characteristics of a region judged as a defect must deviate from the benchmark by more than 3 times the standard deviation, corresponding to a Mahalanobis distance of approximately 9.0, in order to identify marginal outliers within the normal process fluctuation range, a tolerance interval greater than the lower limit of defect judgment needs to be set; in this embodiment, the distance squared corresponding to 4 times the standard deviation is selected as the threshold, i.e. ; like If the cluster is determined to be a marginal false positive, meaning that although the features of the region exceed the old threshold, there has been no statistical qualitative change, then a threshold relaxation operation is performed to allow... Update the current threshold; like If the cluster is determined to be a substantial physical defect, then the threshold update should be forcibly blocked or only the threshold tightening operation should be performed to maintain the quality baseline. Under the premise of satisfying the above discrimination logic and safety boundary constraints, the current operating thresholds of the system are updated using an exponential moving average strategy, as shown in the following formula: , in To update the step size coefficient, for example, 0.05, represent or ; This embodiment introduces an envelope update strategy based on clustering statistical characteristics and combines it with Mahalanobis distance that has undergone strict dimensional normalization to perform binary hypothesis testing on the nature of defects. This avoids the logical flaw that simple mean tracking may lead to threshold involution, i.e., the threshold converges to the defect center, resulting in 50% false negatives. This module enables the system to automatically adjust the detection sensitivity when batch changes in papermaking raw materials cause a shift in the overall optical characteristics, thereby reducing the false alarm rate while maintaining a strict quality baseline.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A paper quality inspection system for myopia prevention based on industrial vision, characterized in that, include: The multi-angle image acquisition module is used to perform grayscale imaging on the surface of anti-myopia paper under multi-angle lighting conditions to obtain grayscale images of the paper surface. The frequency domain analysis module is used to perform local frequency domain transformation on the grayscale image of the paper surface, analyze the frequency distribution characteristics, and calculate the local spectral entropy based on the frequency distribution characteristics to quantify the diffuse reflectance randomness of the microstructure of the paper surface. The texture analysis module is used to perform spatial domain texture statistics on grayscale images of paper surfaces, construct a gray-level co-occurrence matrix, and extract texture energy difference based on the gray-level co-occurrence matrix to characterize the isotropic characteristics of light reflection on the paper surface. The gloss analysis module is used to perform statistical analysis on the pixel grayscale distribution of grayscale images of paper surfaces and calculate the gloss consistency index to evaluate the softness of light on the paper surface under simulated human eye perception. The defect identification module is used to fuse local spectral entropy, texture energy difference, and gloss consistency index to construct a multi-dimensional texture feature set, and to identify functional diffuse reflection abnormal areas based on the multi-dimensional texture feature set to generate quality inspection results.

2. The anti-myopia paper quality inspection system based on industrial vision according to claim 1, characterized in that, Methods for obtaining grayscale images of paper surfaces include: The lighting unit is controlled to illuminate the paper surface sequentially at multiple preset incident angles; At each incident angle, the imaging unit is triggered to expose, and multiple frames of raw illumination images are acquired. Weighted fusion processing is performed on multiple frames of original illumination images to suppress high-frequency random noise and generate a grayscale image of the paper surface.

3. The anti-myopia paper quality inspection system based on industrial vision according to claim 2, characterized in that, Methods for calculating local spectral entropy include: The grayscale image of the paper surface is divided into multiple overlapping local analysis windows; Perform a two-dimensional discrete Fourier transform on the image data within each local analysis window to obtain the local spectral amplitude matrix; The local spectral amplitude matrix is ​​normalized to obtain the normalized spectral probability distribution; Based on the normalized spectral probability distribution, the information entropy value is calculated, and the information entropy value is used as the local spectral entropy corresponding to the local analysis window.

4. The anti-myopia paper quality inspection system based on industrial vision according to claim 3, characterized in that, Methods for extracting texture energy differences include: Based on the grayscale image of the paper surface, multiple pre-selected directional offsets are set; For each directional offset, a corresponding gray-level co-occurrence matrix is ​​constructed. For each gray-level co-occurrence matrix, calculate the quadratic angular moment eigenvalues ​​as the texture energy values ​​for the corresponding directions; Calculate the standard deviation of the texture energy values ​​in all directions, and use the standard deviation as the texture energy difference.

5. The anti-myopia paper quality inspection system based on industrial vision according to claim 4, characterized in that, Methods for calculating the gloss consistency index include: Extract the pixel grayscale histogram of the grayscale image of the paper surface; Calculate the mean, standard deviation, and skewness coefficient of the pixel grayscale histogram; Based on preset weighting coefficients, the standard deviation and skewness coefficient are weighted and summed to obtain the gloss consistency index, where the skewness coefficient is used to characterize the distribution tendency of local highlight areas.

6. The anti-myopia paper quality inspection system based on industrial vision according to claim 5, characterized in that, Methods for identifying functional diffuse aberration regions include: Generate an entropy mapping map and an energy difference mapping map with the same size as the grayscale image of the paper surface; Traverse the entropy value mapping map and filter out areas where the local spectral entropy is lower than the preset diffuse reflection threshold, marking them as suspected specular reflection areas; Traverse the energy difference map and filter out areas where the texture energy difference is higher than the preset isotropic threshold, marking them as suspected coating unevenness areas; The intersection of the suspected specular reflection area and the suspected coating unevenness area is taken, and the intersection area is regarded as the functional diffuse reflection anomaly area.

7. The anti-myopia paper quality inspection system based on industrial vision according to claim 6, characterized in that, Methods for generating quality inspection results include: The percentage area of ​​functional diffuse reflectance anomalies in grayscale images of paper surfaces; Obtain the average gloss consistency index within the functional diffuse reflection anomaly area; A preset quality weight set is provided, which includes area weight and gloss weight. Based on the quality weight set, the visual comfort score is calculated by weighted summation of the area ratio and the average value. The visual comfort score is compared with a preset pass / fail benchmark score: if the visual comfort score is lower than the pass / fail benchmark score, a failure instruction is generated as the quality inspection result; if the visual comfort score is not lower than the pass / fail benchmark score, a pass instruction is generated as the quality inspection result.

8. The anti-myopia paper quality inspection system based on industrial vision according to claim 7, characterized in that, It also includes an adaptive feedback adjustment module, used for: Obtain feature data of functional diffuse reflection anomaly regions corresponding to all non-compliant instructions generated within a historical time period; Cluster analysis was performed on the feature data to identify the cluster centers of the main defect types; Based on the data distribution characteristics of the cluster centers, the preset diffuse reflection threshold and preset isotropic threshold are updated to correct the system's sensitivity to specific process fluctuations.

Images